uipc_usrreq.c revision 252288
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004-2009 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34/* 35 * UNIX Domain (Local) Sockets 36 * 37 * This is an implementation of UNIX (local) domain sockets. Each socket has 38 * an associated struct unpcb (UNIX protocol control block). Stream sockets 39 * may be connected to 0 or 1 other socket. Datagram sockets may be 40 * connected to 0, 1, or many other sockets. Sockets may be created and 41 * connected in pairs (socketpair(2)), or bound/connected to using the file 42 * system name space. For most purposes, only the receive socket buffer is 43 * used, as sending on one socket delivers directly to the receive socket 44 * buffer of a second socket. 45 * 46 * The implementation is substantially complicated by the fact that 47 * "ancillary data", such as file descriptors or credentials, may be passed 48 * across UNIX domain sockets. The potential for passing UNIX domain sockets 49 * over other UNIX domain sockets requires the implementation of a simple 50 * garbage collector to find and tear down cycles of disconnected sockets. 51 * 52 * TODO: 53 * RDM 54 * distinguish datagram size limits from flow control limits in SEQPACKET 55 * rethink name space problems 56 * need a proper out-of-band 57 */ 58 59#include <sys/cdefs.h> 60__FBSDID("$FreeBSD: stable/9/sys/kern/uipc_usrreq.c 252288 2013-06-27 02:27:13Z mjg $"); 61 62#include "opt_ddb.h" 63 64#include <sys/param.h> 65#include <sys/domain.h> 66#include <sys/fcntl.h> 67#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 68#include <sys/eventhandler.h> 69#include <sys/file.h> 70#include <sys/filedesc.h> 71#include <sys/kernel.h> 72#include <sys/lock.h> 73#include <sys/mbuf.h> 74#include <sys/mount.h> 75#include <sys/mutex.h> 76#include <sys/namei.h> 77#include <sys/proc.h> 78#include <sys/protosw.h> 79#include <sys/queue.h> 80#include <sys/resourcevar.h> 81#include <sys/rwlock.h> 82#include <sys/socket.h> 83#include <sys/socketvar.h> 84#include <sys/signalvar.h> 85#include <sys/stat.h> 86#include <sys/sx.h> 87#include <sys/sysctl.h> 88#include <sys/systm.h> 89#include <sys/taskqueue.h> 90#include <sys/un.h> 91#include <sys/unpcb.h> 92#include <sys/vnode.h> 93 94#include <net/vnet.h> 95 96#ifdef DDB 97#include <ddb/ddb.h> 98#endif 99 100#include <security/mac/mac_framework.h> 101 102#include <vm/uma.h> 103 104/* 105 * Locking key: 106 * (l) Locked using list lock 107 * (g) Locked using linkage lock 108 */ 109 110static uma_zone_t unp_zone; 111static unp_gen_t unp_gencnt; /* (l) */ 112static u_int unp_count; /* (l) Count of local sockets. */ 113static ino_t unp_ino; /* Prototype for fake inode numbers. */ 114static int unp_rights; /* (g) File descriptors in flight. */ 115static struct unp_head unp_shead; /* (l) List of stream sockets. */ 116static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 117static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 118 119struct unp_defer { 120 SLIST_ENTRY(unp_defer) ud_link; 121 struct file *ud_fp; 122}; 123static SLIST_HEAD(, unp_defer) unp_defers; 124static int unp_defers_count; 125 126static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 127 128/* 129 * Garbage collection of cyclic file descriptor/socket references occurs 130 * asynchronously in a taskqueue context in order to avoid recursion and 131 * reentrance in the UNIX domain socket, file descriptor, and socket layer 132 * code. See unp_gc() for a full description. 133 */ 134static struct timeout_task unp_gc_task; 135 136/* 137 * The close of unix domain sockets attached as SCM_RIGHTS is 138 * postponed to the taskqueue, to avoid arbitrary recursion depth. 139 * The attached sockets might have another sockets attached. 140 */ 141static struct task unp_defer_task; 142 143/* 144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 145 * stream sockets, although the total for sender and receiver is actually 146 * only PIPSIZ. 147 * 148 * Datagram sockets really use the sendspace as the maximum datagram size, 149 * and don't really want to reserve the sendspace. Their recvspace should be 150 * large enough for at least one max-size datagram plus address. 151 */ 152#ifndef PIPSIZ 153#define PIPSIZ 8192 154#endif 155static u_long unpst_sendspace = PIPSIZ; 156static u_long unpst_recvspace = PIPSIZ; 157static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 158static u_long unpdg_recvspace = 4*1024; 159static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 160static u_long unpsp_recvspace = PIPSIZ; 161 162static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 163static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, 164 "SOCK_STREAM"); 165static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 166static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, 167 "SOCK_SEQPACKET"); 168 169SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 170 &unpst_sendspace, 0, "Default stream send space."); 171SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 172 &unpst_recvspace, 0, "Default stream receive space."); 173SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 174 &unpdg_sendspace, 0, "Default datagram send space."); 175SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 176 &unpdg_recvspace, 0, "Default datagram receive space."); 177SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 178 &unpsp_sendspace, 0, "Default seqpacket send space."); 179SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 180 &unpsp_recvspace, 0, "Default seqpacket receive space."); 181SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 182 "File descriptors in flight."); 183SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 184 &unp_defers_count, 0, 185 "File descriptors deferred to taskqueue for close."); 186 187/* 188 * Locking and synchronization: 189 * 190 * Three types of locks exit in the local domain socket implementation: a 191 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 192 * global locks, the list lock protects the socket count, global generation 193 * number, and stream/datagram global lists. The linkage lock protects the 194 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 195 * held exclusively over the acquisition of multiple unpcb locks to prevent 196 * deadlock. 197 * 198 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 199 * allocated in pru_attach() and freed in pru_detach(). The validity of that 200 * pointer is an invariant, so no lock is required to dereference the so_pcb 201 * pointer if a valid socket reference is held by the caller. In practice, 202 * this is always true during operations performed on a socket. Each unpcb 203 * has a back-pointer to its socket, unp_socket, which will be stable under 204 * the same circumstances. 205 * 206 * This pointer may only be safely dereferenced as long as a valid reference 207 * to the unpcb is held. Typically, this reference will be from the socket, 208 * or from another unpcb when the referring unpcb's lock is held (in order 209 * that the reference not be invalidated during use). For example, to follow 210 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 211 * as unp_socket remains valid as long as the reference to unp_conn is valid. 212 * 213 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 214 * atomic reads without the lock may be performed "lockless", but more 215 * complex reads and read-modify-writes require the mutex to be held. No 216 * lock order is defined between unpcb locks -- multiple unpcb locks may be 217 * acquired at the same time only when holding the linkage rwlock 218 * exclusively, which prevents deadlocks. 219 * 220 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 221 * protocols, bind() is a non-atomic operation, and connect() requires 222 * potential sleeping in the protocol, due to potentially waiting on local or 223 * distributed file systems. We try to separate "lookup" operations, which 224 * may sleep, and the IPC operations themselves, which typically can occur 225 * with relative atomicity as locks can be held over the entire operation. 226 * 227 * Another tricky issue is simultaneous multi-threaded or multi-process 228 * access to a single UNIX domain socket. These are handled by the flags 229 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 230 * binding, both of which involve dropping UNIX domain socket locks in order 231 * to perform namei() and other file system operations. 232 */ 233static struct rwlock unp_link_rwlock; 234static struct mtx unp_list_lock; 235static struct mtx unp_defers_lock; 236 237#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 238 "unp_link_rwlock") 239 240#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 241 RA_LOCKED) 242#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 243 RA_UNLOCKED) 244 245#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 246#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 247#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 248#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 249#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 250 RA_WLOCKED) 251 252#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 253 "unp_list_lock", NULL, MTX_DEF) 254#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 255#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 256 257#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 258 "unp_defer", NULL, MTX_DEF) 259#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 260#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 261 262#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 263 "unp_mtx", "unp_mtx", \ 264 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 265#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 266#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 267#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 268#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 269 270static int uipc_connect2(struct socket *, struct socket *); 271static int uipc_ctloutput(struct socket *, struct sockopt *); 272static int unp_connect(struct socket *, struct sockaddr *, 273 struct thread *); 274static int unp_connect2(struct socket *so, struct socket *so2, int); 275static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 276static void unp_dispose(struct mbuf *); 277static void unp_shutdown(struct unpcb *); 278static void unp_drop(struct unpcb *, int); 279static void unp_gc(__unused void *, int); 280static void unp_scan(struct mbuf *, void (*)(struct file *)); 281static void unp_discard(struct file *); 282static void unp_freerights(struct file **, int); 283static void unp_init(void); 284static int unp_internalize(struct mbuf **, struct thread *); 285static void unp_internalize_fp(struct file *); 286static int unp_externalize(struct mbuf *, struct mbuf **); 287static int unp_externalize_fp(struct file *); 288static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 289static void unp_process_defers(void * __unused, int); 290 291/* 292 * Definitions of protocols supported in the LOCAL domain. 293 */ 294static struct domain localdomain; 295static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 296static struct pr_usrreqs uipc_usrreqs_seqpacket; 297static struct protosw localsw[] = { 298{ 299 .pr_type = SOCK_STREAM, 300 .pr_domain = &localdomain, 301 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 302 .pr_ctloutput = &uipc_ctloutput, 303 .pr_usrreqs = &uipc_usrreqs_stream 304}, 305{ 306 .pr_type = SOCK_DGRAM, 307 .pr_domain = &localdomain, 308 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 309 .pr_ctloutput = &uipc_ctloutput, 310 .pr_usrreqs = &uipc_usrreqs_dgram 311}, 312{ 313 .pr_type = SOCK_SEQPACKET, 314 .pr_domain = &localdomain, 315 316 /* 317 * XXXRW: For now, PR_ADDR because soreceive will bump into them 318 * due to our use of sbappendaddr. A new sbappend variants is needed 319 * that supports both atomic record writes and control data. 320 */ 321 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 322 PR_RIGHTS, 323 .pr_usrreqs = &uipc_usrreqs_seqpacket, 324}, 325}; 326 327static struct domain localdomain = { 328 .dom_family = AF_LOCAL, 329 .dom_name = "local", 330 .dom_init = unp_init, 331 .dom_externalize = unp_externalize, 332 .dom_dispose = unp_dispose, 333 .dom_protosw = localsw, 334 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 335}; 336DOMAIN_SET(local); 337 338static void 339uipc_abort(struct socket *so) 340{ 341 struct unpcb *unp, *unp2; 342 343 unp = sotounpcb(so); 344 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 345 346 UNP_LINK_WLOCK(); 347 UNP_PCB_LOCK(unp); 348 unp2 = unp->unp_conn; 349 if (unp2 != NULL) { 350 UNP_PCB_LOCK(unp2); 351 unp_drop(unp2, ECONNABORTED); 352 UNP_PCB_UNLOCK(unp2); 353 } 354 UNP_PCB_UNLOCK(unp); 355 UNP_LINK_WUNLOCK(); 356} 357 358static int 359uipc_accept(struct socket *so, struct sockaddr **nam) 360{ 361 struct unpcb *unp, *unp2; 362 const struct sockaddr *sa; 363 364 /* 365 * Pass back name of connected socket, if it was bound and we are 366 * still connected (our peer may have closed already!). 367 */ 368 unp = sotounpcb(so); 369 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 370 371 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 372 UNP_LINK_RLOCK(); 373 unp2 = unp->unp_conn; 374 if (unp2 != NULL && unp2->unp_addr != NULL) { 375 UNP_PCB_LOCK(unp2); 376 sa = (struct sockaddr *) unp2->unp_addr; 377 bcopy(sa, *nam, sa->sa_len); 378 UNP_PCB_UNLOCK(unp2); 379 } else { 380 sa = &sun_noname; 381 bcopy(sa, *nam, sa->sa_len); 382 } 383 UNP_LINK_RUNLOCK(); 384 return (0); 385} 386 387static int 388uipc_attach(struct socket *so, int proto, struct thread *td) 389{ 390 u_long sendspace, recvspace; 391 struct unpcb *unp; 392 int error; 393 394 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 395 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 396 switch (so->so_type) { 397 case SOCK_STREAM: 398 sendspace = unpst_sendspace; 399 recvspace = unpst_recvspace; 400 break; 401 402 case SOCK_DGRAM: 403 sendspace = unpdg_sendspace; 404 recvspace = unpdg_recvspace; 405 break; 406 407 case SOCK_SEQPACKET: 408 sendspace = unpsp_sendspace; 409 recvspace = unpsp_recvspace; 410 break; 411 412 default: 413 panic("uipc_attach"); 414 } 415 error = soreserve(so, sendspace, recvspace); 416 if (error) 417 return (error); 418 } 419 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 420 if (unp == NULL) 421 return (ENOBUFS); 422 LIST_INIT(&unp->unp_refs); 423 UNP_PCB_LOCK_INIT(unp); 424 unp->unp_socket = so; 425 so->so_pcb = unp; 426 unp->unp_refcount = 1; 427 428 UNP_LIST_LOCK(); 429 unp->unp_gencnt = ++unp_gencnt; 430 unp_count++; 431 switch (so->so_type) { 432 case SOCK_STREAM: 433 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 434 break; 435 436 case SOCK_DGRAM: 437 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 438 break; 439 440 case SOCK_SEQPACKET: 441 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 442 break; 443 444 default: 445 panic("uipc_attach"); 446 } 447 UNP_LIST_UNLOCK(); 448 449 return (0); 450} 451 452static int 453uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 454{ 455 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 456 struct vattr vattr; 457 int error, namelen, vfslocked; 458 struct nameidata nd; 459 struct unpcb *unp; 460 struct vnode *vp; 461 struct mount *mp; 462 char *buf; 463 464 unp = sotounpcb(so); 465 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 466 467 if (soun->sun_len > sizeof(struct sockaddr_un)) 468 return (EINVAL); 469 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 470 if (namelen <= 0) 471 return (EINVAL); 472 473 /* 474 * We don't allow simultaneous bind() calls on a single UNIX domain 475 * socket, so flag in-progress operations, and return an error if an 476 * operation is already in progress. 477 * 478 * Historically, we have not allowed a socket to be rebound, so this 479 * also returns an error. Not allowing re-binding simplifies the 480 * implementation and avoids a great many possible failure modes. 481 */ 482 UNP_PCB_LOCK(unp); 483 if (unp->unp_vnode != NULL) { 484 UNP_PCB_UNLOCK(unp); 485 return (EINVAL); 486 } 487 if (unp->unp_flags & UNP_BINDING) { 488 UNP_PCB_UNLOCK(unp); 489 return (EALREADY); 490 } 491 unp->unp_flags |= UNP_BINDING; 492 UNP_PCB_UNLOCK(unp); 493 494 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 495 bcopy(soun->sun_path, buf, namelen); 496 buf[namelen] = 0; 497 498restart: 499 vfslocked = 0; 500 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 501 UIO_SYSSPACE, buf, td); 502/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 503 error = namei(&nd); 504 if (error) 505 goto error; 506 vp = nd.ni_vp; 507 vfslocked = NDHASGIANT(&nd); 508 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 509 NDFREE(&nd, NDF_ONLY_PNBUF); 510 if (nd.ni_dvp == vp) 511 vrele(nd.ni_dvp); 512 else 513 vput(nd.ni_dvp); 514 if (vp != NULL) { 515 vrele(vp); 516 error = EADDRINUSE; 517 goto error; 518 } 519 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 520 if (error) 521 goto error; 522 VFS_UNLOCK_GIANT(vfslocked); 523 goto restart; 524 } 525 VATTR_NULL(&vattr); 526 vattr.va_type = VSOCK; 527 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 528#ifdef MAC 529 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 530 &vattr); 531#endif 532 if (error == 0) 533 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 534 NDFREE(&nd, NDF_ONLY_PNBUF); 535 vput(nd.ni_dvp); 536 if (error) { 537 vn_finished_write(mp); 538 goto error; 539 } 540 vp = nd.ni_vp; 541 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 542 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 543 544 UNP_LINK_WLOCK(); 545 UNP_PCB_LOCK(unp); 546 VOP_UNP_BIND(vp, unp->unp_socket); 547 unp->unp_vnode = vp; 548 unp->unp_addr = soun; 549 unp->unp_flags &= ~UNP_BINDING; 550 UNP_PCB_UNLOCK(unp); 551 UNP_LINK_WUNLOCK(); 552 VOP_UNLOCK(vp, 0); 553 vn_finished_write(mp); 554 VFS_UNLOCK_GIANT(vfslocked); 555 free(buf, M_TEMP); 556 return (0); 557 558error: 559 VFS_UNLOCK_GIANT(vfslocked); 560 UNP_PCB_LOCK(unp); 561 unp->unp_flags &= ~UNP_BINDING; 562 UNP_PCB_UNLOCK(unp); 563 free(buf, M_TEMP); 564 return (error); 565} 566 567static int 568uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 569{ 570 int error; 571 572 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 573 UNP_LINK_WLOCK(); 574 error = unp_connect(so, nam, td); 575 UNP_LINK_WUNLOCK(); 576 return (error); 577} 578 579static void 580uipc_close(struct socket *so) 581{ 582 struct unpcb *unp, *unp2; 583 584 unp = sotounpcb(so); 585 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 586 587 UNP_LINK_WLOCK(); 588 UNP_PCB_LOCK(unp); 589 unp2 = unp->unp_conn; 590 if (unp2 != NULL) { 591 UNP_PCB_LOCK(unp2); 592 unp_disconnect(unp, unp2); 593 UNP_PCB_UNLOCK(unp2); 594 } 595 UNP_PCB_UNLOCK(unp); 596 UNP_LINK_WUNLOCK(); 597} 598 599static int 600uipc_connect2(struct socket *so1, struct socket *so2) 601{ 602 struct unpcb *unp, *unp2; 603 int error; 604 605 UNP_LINK_WLOCK(); 606 unp = so1->so_pcb; 607 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 608 UNP_PCB_LOCK(unp); 609 unp2 = so2->so_pcb; 610 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 611 UNP_PCB_LOCK(unp2); 612 error = unp_connect2(so1, so2, PRU_CONNECT2); 613 UNP_PCB_UNLOCK(unp2); 614 UNP_PCB_UNLOCK(unp); 615 UNP_LINK_WUNLOCK(); 616 return (error); 617} 618 619static void 620uipc_detach(struct socket *so) 621{ 622 struct unpcb *unp, *unp2; 623 struct sockaddr_un *saved_unp_addr; 624 struct vnode *vp; 625 int freeunp, local_unp_rights; 626 627 unp = sotounpcb(so); 628 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 629 630 UNP_LINK_WLOCK(); 631 UNP_LIST_LOCK(); 632 UNP_PCB_LOCK(unp); 633 LIST_REMOVE(unp, unp_link); 634 unp->unp_gencnt = ++unp_gencnt; 635 --unp_count; 636 UNP_LIST_UNLOCK(); 637 638 /* 639 * XXXRW: Should assert vp->v_socket == so. 640 */ 641 if ((vp = unp->unp_vnode) != NULL) { 642 VOP_UNP_DETACH(vp); 643 unp->unp_vnode = NULL; 644 } 645 unp2 = unp->unp_conn; 646 if (unp2 != NULL) { 647 UNP_PCB_LOCK(unp2); 648 unp_disconnect(unp, unp2); 649 UNP_PCB_UNLOCK(unp2); 650 } 651 652 /* 653 * We hold the linkage lock exclusively, so it's OK to acquire 654 * multiple pcb locks at a time. 655 */ 656 while (!LIST_EMPTY(&unp->unp_refs)) { 657 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 658 659 UNP_PCB_LOCK(ref); 660 unp_drop(ref, ECONNRESET); 661 UNP_PCB_UNLOCK(ref); 662 } 663 local_unp_rights = unp_rights; 664 UNP_LINK_WUNLOCK(); 665 unp->unp_socket->so_pcb = NULL; 666 saved_unp_addr = unp->unp_addr; 667 unp->unp_addr = NULL; 668 unp->unp_refcount--; 669 freeunp = (unp->unp_refcount == 0); 670 if (saved_unp_addr != NULL) 671 free(saved_unp_addr, M_SONAME); 672 if (freeunp) { 673 UNP_PCB_LOCK_DESTROY(unp); 674 uma_zfree(unp_zone, unp); 675 } else 676 UNP_PCB_UNLOCK(unp); 677 if (vp) { 678 int vfslocked; 679 680 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 681 vrele(vp); 682 VFS_UNLOCK_GIANT(vfslocked); 683 } 684 if (local_unp_rights) 685 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 686} 687 688static int 689uipc_disconnect(struct socket *so) 690{ 691 struct unpcb *unp, *unp2; 692 693 unp = sotounpcb(so); 694 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 695 696 UNP_LINK_WLOCK(); 697 UNP_PCB_LOCK(unp); 698 unp2 = unp->unp_conn; 699 if (unp2 != NULL) { 700 UNP_PCB_LOCK(unp2); 701 unp_disconnect(unp, unp2); 702 UNP_PCB_UNLOCK(unp2); 703 } 704 UNP_PCB_UNLOCK(unp); 705 UNP_LINK_WUNLOCK(); 706 return (0); 707} 708 709static int 710uipc_listen(struct socket *so, int backlog, struct thread *td) 711{ 712 struct unpcb *unp; 713 int error; 714 715 unp = sotounpcb(so); 716 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 717 718 UNP_PCB_LOCK(unp); 719 if (unp->unp_vnode == NULL) { 720 UNP_PCB_UNLOCK(unp); 721 return (EINVAL); 722 } 723 724 SOCK_LOCK(so); 725 error = solisten_proto_check(so); 726 if (error == 0) { 727 cru2x(td->td_ucred, &unp->unp_peercred); 728 unp->unp_flags |= UNP_HAVEPCCACHED; 729 solisten_proto(so, backlog); 730 } 731 SOCK_UNLOCK(so); 732 UNP_PCB_UNLOCK(unp); 733 return (error); 734} 735 736static int 737uipc_peeraddr(struct socket *so, struct sockaddr **nam) 738{ 739 struct unpcb *unp, *unp2; 740 const struct sockaddr *sa; 741 742 unp = sotounpcb(so); 743 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 744 745 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 746 UNP_LINK_RLOCK(); 747 /* 748 * XXX: It seems that this test always fails even when connection is 749 * established. So, this else clause is added as workaround to 750 * return PF_LOCAL sockaddr. 751 */ 752 unp2 = unp->unp_conn; 753 if (unp2 != NULL) { 754 UNP_PCB_LOCK(unp2); 755 if (unp2->unp_addr != NULL) 756 sa = (struct sockaddr *) unp2->unp_addr; 757 else 758 sa = &sun_noname; 759 bcopy(sa, *nam, sa->sa_len); 760 UNP_PCB_UNLOCK(unp2); 761 } else { 762 sa = &sun_noname; 763 bcopy(sa, *nam, sa->sa_len); 764 } 765 UNP_LINK_RUNLOCK(); 766 return (0); 767} 768 769static int 770uipc_rcvd(struct socket *so, int flags) 771{ 772 struct unpcb *unp, *unp2; 773 struct socket *so2; 774 u_int mbcnt, sbcc; 775 u_long newhiwat; 776 777 unp = sotounpcb(so); 778 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 779 780 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 781 panic("uipc_rcvd socktype %d", so->so_type); 782 783 /* 784 * Adjust backpressure on sender and wakeup any waiting to write. 785 * 786 * The unp lock is acquired to maintain the validity of the unp_conn 787 * pointer; no lock on unp2 is required as unp2->unp_socket will be 788 * static as long as we don't permit unp2 to disconnect from unp, 789 * which is prevented by the lock on unp. We cache values from 790 * so_rcv to avoid holding the so_rcv lock over the entire 791 * transaction on the remote so_snd. 792 */ 793 SOCKBUF_LOCK(&so->so_rcv); 794 mbcnt = so->so_rcv.sb_mbcnt; 795 sbcc = so->so_rcv.sb_cc; 796 SOCKBUF_UNLOCK(&so->so_rcv); 797 UNP_PCB_LOCK(unp); 798 unp2 = unp->unp_conn; 799 if (unp2 == NULL) { 800 UNP_PCB_UNLOCK(unp); 801 return (0); 802 } 803 so2 = unp2->unp_socket; 804 SOCKBUF_LOCK(&so2->so_snd); 805 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 806 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 807 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 808 newhiwat, RLIM_INFINITY); 809 sowwakeup_locked(so2); 810 unp->unp_mbcnt = mbcnt; 811 unp->unp_cc = sbcc; 812 UNP_PCB_UNLOCK(unp); 813 return (0); 814} 815 816static int 817uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 818 struct mbuf *control, struct thread *td) 819{ 820 struct unpcb *unp, *unp2; 821 struct socket *so2; 822 u_int mbcnt_delta, sbcc; 823 u_int newhiwat; 824 int error = 0; 825 826 unp = sotounpcb(so); 827 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 828 829 if (flags & PRUS_OOB) { 830 error = EOPNOTSUPP; 831 goto release; 832 } 833 if (control != NULL && (error = unp_internalize(&control, td))) 834 goto release; 835 if ((nam != NULL) || (flags & PRUS_EOF)) 836 UNP_LINK_WLOCK(); 837 else 838 UNP_LINK_RLOCK(); 839 switch (so->so_type) { 840 case SOCK_DGRAM: 841 { 842 const struct sockaddr *from; 843 844 unp2 = unp->unp_conn; 845 if (nam != NULL) { 846 UNP_LINK_WLOCK_ASSERT(); 847 if (unp2 != NULL) { 848 error = EISCONN; 849 break; 850 } 851 error = unp_connect(so, nam, td); 852 if (error) 853 break; 854 unp2 = unp->unp_conn; 855 } 856 857 /* 858 * Because connect() and send() are non-atomic in a sendto() 859 * with a target address, it's possible that the socket will 860 * have disconnected before the send() can run. In that case 861 * return the slightly counter-intuitive but otherwise 862 * correct error that the socket is not connected. 863 */ 864 if (unp2 == NULL) { 865 error = ENOTCONN; 866 break; 867 } 868 /* Lockless read. */ 869 if (unp2->unp_flags & UNP_WANTCRED) 870 control = unp_addsockcred(td, control); 871 UNP_PCB_LOCK(unp); 872 if (unp->unp_addr != NULL) 873 from = (struct sockaddr *)unp->unp_addr; 874 else 875 from = &sun_noname; 876 so2 = unp2->unp_socket; 877 SOCKBUF_LOCK(&so2->so_rcv); 878 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 879 sorwakeup_locked(so2); 880 m = NULL; 881 control = NULL; 882 } else { 883 SOCKBUF_UNLOCK(&so2->so_rcv); 884 error = ENOBUFS; 885 } 886 if (nam != NULL) { 887 UNP_LINK_WLOCK_ASSERT(); 888 UNP_PCB_LOCK(unp2); 889 unp_disconnect(unp, unp2); 890 UNP_PCB_UNLOCK(unp2); 891 } 892 UNP_PCB_UNLOCK(unp); 893 break; 894 } 895 896 case SOCK_SEQPACKET: 897 case SOCK_STREAM: 898 if ((so->so_state & SS_ISCONNECTED) == 0) { 899 if (nam != NULL) { 900 UNP_LINK_WLOCK_ASSERT(); 901 error = unp_connect(so, nam, td); 902 if (error) 903 break; /* XXX */ 904 } else { 905 error = ENOTCONN; 906 break; 907 } 908 } 909 910 /* Lockless read. */ 911 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 912 error = EPIPE; 913 break; 914 } 915 916 /* 917 * Because connect() and send() are non-atomic in a sendto() 918 * with a target address, it's possible that the socket will 919 * have disconnected before the send() can run. In that case 920 * return the slightly counter-intuitive but otherwise 921 * correct error that the socket is not connected. 922 * 923 * Locking here must be done carefully: the linkage lock 924 * prevents interconnections between unpcbs from changing, so 925 * we can traverse from unp to unp2 without acquiring unp's 926 * lock. Socket buffer locks follow unpcb locks, so we can 927 * acquire both remote and lock socket buffer locks. 928 */ 929 unp2 = unp->unp_conn; 930 if (unp2 == NULL) { 931 error = ENOTCONN; 932 break; 933 } 934 so2 = unp2->unp_socket; 935 UNP_PCB_LOCK(unp2); 936 SOCKBUF_LOCK(&so2->so_rcv); 937 if (unp2->unp_flags & UNP_WANTCRED) { 938 /* 939 * Credentials are passed only once on SOCK_STREAM. 940 */ 941 unp2->unp_flags &= ~UNP_WANTCRED; 942 control = unp_addsockcred(td, control); 943 } 944 /* 945 * Send to paired receive port, and then reduce send buffer 946 * hiwater marks to maintain backpressure. Wake up readers. 947 */ 948 switch (so->so_type) { 949 case SOCK_STREAM: 950 if (control != NULL) { 951 if (sbappendcontrol_locked(&so2->so_rcv, m, 952 control)) 953 control = NULL; 954 } else 955 sbappend_locked(&so2->so_rcv, m); 956 break; 957 958 case SOCK_SEQPACKET: { 959 const struct sockaddr *from; 960 961 from = &sun_noname; 962 if (sbappendaddr_locked(&so2->so_rcv, from, m, 963 control)) 964 control = NULL; 965 break; 966 } 967 } 968 969 /* 970 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 971 * datagram size and back-pressure for SOCK_SEQPACKET, which 972 * can lead to undesired return of EMSGSIZE on send instead 973 * of more desirable blocking. 974 */ 975 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 976 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 977 sbcc = so2->so_rcv.sb_cc; 978 sorwakeup_locked(so2); 979 980 SOCKBUF_LOCK(&so->so_snd); 981 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 982 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 983 else 984 newhiwat = 0; 985 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 986 newhiwat, RLIM_INFINITY); 987 so->so_snd.sb_mbmax -= mbcnt_delta; 988 SOCKBUF_UNLOCK(&so->so_snd); 989 unp2->unp_cc = sbcc; 990 UNP_PCB_UNLOCK(unp2); 991 m = NULL; 992 break; 993 994 default: 995 panic("uipc_send unknown socktype"); 996 } 997 998 /* 999 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1000 */ 1001 if (flags & PRUS_EOF) { 1002 UNP_PCB_LOCK(unp); 1003 socantsendmore(so); 1004 unp_shutdown(unp); 1005 UNP_PCB_UNLOCK(unp); 1006 } 1007 1008 if ((nam != NULL) || (flags & PRUS_EOF)) 1009 UNP_LINK_WUNLOCK(); 1010 else 1011 UNP_LINK_RUNLOCK(); 1012 1013 if (control != NULL && error != 0) 1014 unp_dispose(control); 1015 1016release: 1017 if (control != NULL) 1018 m_freem(control); 1019 if (m != NULL) 1020 m_freem(m); 1021 return (error); 1022} 1023 1024static int 1025uipc_sense(struct socket *so, struct stat *sb) 1026{ 1027 struct unpcb *unp, *unp2; 1028 struct socket *so2; 1029 1030 unp = sotounpcb(so); 1031 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1032 1033 sb->st_blksize = so->so_snd.sb_hiwat; 1034 UNP_LINK_RLOCK(); 1035 UNP_PCB_LOCK(unp); 1036 unp2 = unp->unp_conn; 1037 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1038 unp2 != NULL) { 1039 so2 = unp2->unp_socket; 1040 sb->st_blksize += so2->so_rcv.sb_cc; 1041 } 1042 sb->st_dev = NODEV; 1043 if (unp->unp_ino == 0) 1044 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1045 sb->st_ino = unp->unp_ino; 1046 UNP_PCB_UNLOCK(unp); 1047 UNP_LINK_RUNLOCK(); 1048 return (0); 1049} 1050 1051static int 1052uipc_shutdown(struct socket *so) 1053{ 1054 struct unpcb *unp; 1055 1056 unp = sotounpcb(so); 1057 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1058 1059 UNP_LINK_WLOCK(); 1060 UNP_PCB_LOCK(unp); 1061 socantsendmore(so); 1062 unp_shutdown(unp); 1063 UNP_PCB_UNLOCK(unp); 1064 UNP_LINK_WUNLOCK(); 1065 return (0); 1066} 1067 1068static int 1069uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1070{ 1071 struct unpcb *unp; 1072 const struct sockaddr *sa; 1073 1074 unp = sotounpcb(so); 1075 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1076 1077 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1078 UNP_PCB_LOCK(unp); 1079 if (unp->unp_addr != NULL) 1080 sa = (struct sockaddr *) unp->unp_addr; 1081 else 1082 sa = &sun_noname; 1083 bcopy(sa, *nam, sa->sa_len); 1084 UNP_PCB_UNLOCK(unp); 1085 return (0); 1086} 1087 1088static struct pr_usrreqs uipc_usrreqs_dgram = { 1089 .pru_abort = uipc_abort, 1090 .pru_accept = uipc_accept, 1091 .pru_attach = uipc_attach, 1092 .pru_bind = uipc_bind, 1093 .pru_connect = uipc_connect, 1094 .pru_connect2 = uipc_connect2, 1095 .pru_detach = uipc_detach, 1096 .pru_disconnect = uipc_disconnect, 1097 .pru_listen = uipc_listen, 1098 .pru_peeraddr = uipc_peeraddr, 1099 .pru_rcvd = uipc_rcvd, 1100 .pru_send = uipc_send, 1101 .pru_sense = uipc_sense, 1102 .pru_shutdown = uipc_shutdown, 1103 .pru_sockaddr = uipc_sockaddr, 1104 .pru_soreceive = soreceive_dgram, 1105 .pru_close = uipc_close, 1106}; 1107 1108static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1109 .pru_abort = uipc_abort, 1110 .pru_accept = uipc_accept, 1111 .pru_attach = uipc_attach, 1112 .pru_bind = uipc_bind, 1113 .pru_connect = uipc_connect, 1114 .pru_connect2 = uipc_connect2, 1115 .pru_detach = uipc_detach, 1116 .pru_disconnect = uipc_disconnect, 1117 .pru_listen = uipc_listen, 1118 .pru_peeraddr = uipc_peeraddr, 1119 .pru_rcvd = uipc_rcvd, 1120 .pru_send = uipc_send, 1121 .pru_sense = uipc_sense, 1122 .pru_shutdown = uipc_shutdown, 1123 .pru_sockaddr = uipc_sockaddr, 1124 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1125 .pru_close = uipc_close, 1126}; 1127 1128static struct pr_usrreqs uipc_usrreqs_stream = { 1129 .pru_abort = uipc_abort, 1130 .pru_accept = uipc_accept, 1131 .pru_attach = uipc_attach, 1132 .pru_bind = uipc_bind, 1133 .pru_connect = uipc_connect, 1134 .pru_connect2 = uipc_connect2, 1135 .pru_detach = uipc_detach, 1136 .pru_disconnect = uipc_disconnect, 1137 .pru_listen = uipc_listen, 1138 .pru_peeraddr = uipc_peeraddr, 1139 .pru_rcvd = uipc_rcvd, 1140 .pru_send = uipc_send, 1141 .pru_sense = uipc_sense, 1142 .pru_shutdown = uipc_shutdown, 1143 .pru_sockaddr = uipc_sockaddr, 1144 .pru_soreceive = soreceive_generic, 1145 .pru_close = uipc_close, 1146}; 1147 1148static int 1149uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1150{ 1151 struct unpcb *unp; 1152 struct xucred xu; 1153 int error, optval; 1154 1155 if (sopt->sopt_level != 0) 1156 return (EINVAL); 1157 1158 unp = sotounpcb(so); 1159 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1160 error = 0; 1161 switch (sopt->sopt_dir) { 1162 case SOPT_GET: 1163 switch (sopt->sopt_name) { 1164 case LOCAL_PEERCRED: 1165 UNP_PCB_LOCK(unp); 1166 if (unp->unp_flags & UNP_HAVEPC) 1167 xu = unp->unp_peercred; 1168 else { 1169 if (so->so_type == SOCK_STREAM) 1170 error = ENOTCONN; 1171 else 1172 error = EINVAL; 1173 } 1174 UNP_PCB_UNLOCK(unp); 1175 if (error == 0) 1176 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1177 break; 1178 1179 case LOCAL_CREDS: 1180 /* Unlocked read. */ 1181 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1182 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1183 break; 1184 1185 case LOCAL_CONNWAIT: 1186 /* Unlocked read. */ 1187 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1188 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1189 break; 1190 1191 default: 1192 error = EOPNOTSUPP; 1193 break; 1194 } 1195 break; 1196 1197 case SOPT_SET: 1198 switch (sopt->sopt_name) { 1199 case LOCAL_CREDS: 1200 case LOCAL_CONNWAIT: 1201 error = sooptcopyin(sopt, &optval, sizeof(optval), 1202 sizeof(optval)); 1203 if (error) 1204 break; 1205 1206#define OPTSET(bit) do { \ 1207 UNP_PCB_LOCK(unp); \ 1208 if (optval) \ 1209 unp->unp_flags |= bit; \ 1210 else \ 1211 unp->unp_flags &= ~bit; \ 1212 UNP_PCB_UNLOCK(unp); \ 1213} while (0) 1214 1215 switch (sopt->sopt_name) { 1216 case LOCAL_CREDS: 1217 OPTSET(UNP_WANTCRED); 1218 break; 1219 1220 case LOCAL_CONNWAIT: 1221 OPTSET(UNP_CONNWAIT); 1222 break; 1223 1224 default: 1225 break; 1226 } 1227 break; 1228#undef OPTSET 1229 default: 1230 error = ENOPROTOOPT; 1231 break; 1232 } 1233 break; 1234 1235 default: 1236 error = EOPNOTSUPP; 1237 break; 1238 } 1239 return (error); 1240} 1241 1242static int 1243unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1244{ 1245 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1246 struct vnode *vp; 1247 struct socket *so2, *so3; 1248 struct unpcb *unp, *unp2, *unp3; 1249 int error, len, vfslocked; 1250 struct nameidata nd; 1251 char buf[SOCK_MAXADDRLEN]; 1252 struct sockaddr *sa; 1253 1254 UNP_LINK_WLOCK_ASSERT(); 1255 1256 unp = sotounpcb(so); 1257 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1258 1259 if (nam->sa_len > sizeof(struct sockaddr_un)) 1260 return (EINVAL); 1261 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1262 if (len <= 0) 1263 return (EINVAL); 1264 bcopy(soun->sun_path, buf, len); 1265 buf[len] = 0; 1266 1267 UNP_PCB_LOCK(unp); 1268 if (unp->unp_flags & UNP_CONNECTING) { 1269 UNP_PCB_UNLOCK(unp); 1270 return (EALREADY); 1271 } 1272 UNP_LINK_WUNLOCK(); 1273 unp->unp_flags |= UNP_CONNECTING; 1274 UNP_PCB_UNLOCK(unp); 1275 1276 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1277 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKSHARED | LOCKLEAF, 1278 UIO_SYSSPACE, buf, td); 1279 error = namei(&nd); 1280 if (error) 1281 vp = NULL; 1282 else 1283 vp = nd.ni_vp; 1284 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1285 vfslocked = NDHASGIANT(&nd); 1286 NDFREE(&nd, NDF_ONLY_PNBUF); 1287 if (error) 1288 goto bad; 1289 1290 if (vp->v_type != VSOCK) { 1291 error = ENOTSOCK; 1292 goto bad; 1293 } 1294#ifdef MAC 1295 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1296 if (error) 1297 goto bad; 1298#endif 1299 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1300 if (error) 1301 goto bad; 1302 VFS_UNLOCK_GIANT(vfslocked); 1303 1304 unp = sotounpcb(so); 1305 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1306 1307 /* 1308 * Lock linkage lock for two reasons: make sure v_socket is stable, 1309 * and to protect simultaneous locking of multiple pcbs. 1310 */ 1311 UNP_LINK_WLOCK(); 1312 VOP_UNP_CONNECT(vp, &so2); 1313 if (so2 == NULL) { 1314 error = ECONNREFUSED; 1315 goto bad2; 1316 } 1317 if (so->so_type != so2->so_type) { 1318 error = EPROTOTYPE; 1319 goto bad2; 1320 } 1321 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1322 if (so2->so_options & SO_ACCEPTCONN) { 1323 CURVNET_SET(so2->so_vnet); 1324 so3 = sonewconn(so2, 0); 1325 CURVNET_RESTORE(); 1326 } else 1327 so3 = NULL; 1328 if (so3 == NULL) { 1329 error = ECONNREFUSED; 1330 goto bad2; 1331 } 1332 unp = sotounpcb(so); 1333 unp2 = sotounpcb(so2); 1334 unp3 = sotounpcb(so3); 1335 UNP_PCB_LOCK(unp); 1336 UNP_PCB_LOCK(unp2); 1337 UNP_PCB_LOCK(unp3); 1338 if (unp2->unp_addr != NULL) { 1339 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1340 unp3->unp_addr = (struct sockaddr_un *) sa; 1341 sa = NULL; 1342 } 1343 1344 /* 1345 * The connecter's (client's) credentials are copied from its 1346 * process structure at the time of connect() (which is now). 1347 */ 1348 cru2x(td->td_ucred, &unp3->unp_peercred); 1349 unp3->unp_flags |= UNP_HAVEPC; 1350 1351 /* 1352 * The receiver's (server's) credentials are copied from the 1353 * unp_peercred member of socket on which the former called 1354 * listen(); uipc_listen() cached that process's credentials 1355 * at that time so we can use them now. 1356 */ 1357 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1358 ("unp_connect: listener without cached peercred")); 1359 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1360 sizeof(unp->unp_peercred)); 1361 unp->unp_flags |= UNP_HAVEPC; 1362 if (unp2->unp_flags & UNP_WANTCRED) 1363 unp3->unp_flags |= UNP_WANTCRED; 1364 UNP_PCB_UNLOCK(unp3); 1365 UNP_PCB_UNLOCK(unp2); 1366 UNP_PCB_UNLOCK(unp); 1367#ifdef MAC 1368 mac_socketpeer_set_from_socket(so, so3); 1369 mac_socketpeer_set_from_socket(so3, so); 1370#endif 1371 1372 so2 = so3; 1373 } 1374 unp = sotounpcb(so); 1375 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1376 unp2 = sotounpcb(so2); 1377 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1378 UNP_PCB_LOCK(unp); 1379 UNP_PCB_LOCK(unp2); 1380 error = unp_connect2(so, so2, PRU_CONNECT); 1381 UNP_PCB_UNLOCK(unp2); 1382 UNP_PCB_UNLOCK(unp); 1383bad2: 1384 UNP_LINK_WUNLOCK(); 1385 if (vfslocked) 1386 /* 1387 * Giant has been previously acquired. This means filesystem 1388 * isn't MPSAFE. Do it once again. 1389 */ 1390 mtx_lock(&Giant); 1391bad: 1392 if (vp != NULL) 1393 vput(vp); 1394 VFS_UNLOCK_GIANT(vfslocked); 1395 free(sa, M_SONAME); 1396 UNP_LINK_WLOCK(); 1397 UNP_PCB_LOCK(unp); 1398 unp->unp_flags &= ~UNP_CONNECTING; 1399 UNP_PCB_UNLOCK(unp); 1400 return (error); 1401} 1402 1403static int 1404unp_connect2(struct socket *so, struct socket *so2, int req) 1405{ 1406 struct unpcb *unp; 1407 struct unpcb *unp2; 1408 1409 unp = sotounpcb(so); 1410 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1411 unp2 = sotounpcb(so2); 1412 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1413 1414 UNP_LINK_WLOCK_ASSERT(); 1415 UNP_PCB_LOCK_ASSERT(unp); 1416 UNP_PCB_LOCK_ASSERT(unp2); 1417 1418 if (so2->so_type != so->so_type) 1419 return (EPROTOTYPE); 1420 unp->unp_conn = unp2; 1421 1422 switch (so->so_type) { 1423 case SOCK_DGRAM: 1424 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1425 soisconnected(so); 1426 break; 1427 1428 case SOCK_STREAM: 1429 case SOCK_SEQPACKET: 1430 unp2->unp_conn = unp; 1431 if (req == PRU_CONNECT && 1432 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1433 soisconnecting(so); 1434 else 1435 soisconnected(so); 1436 soisconnected(so2); 1437 break; 1438 1439 default: 1440 panic("unp_connect2"); 1441 } 1442 return (0); 1443} 1444 1445static void 1446unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1447{ 1448 struct socket *so; 1449 1450 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1451 1452 UNP_LINK_WLOCK_ASSERT(); 1453 UNP_PCB_LOCK_ASSERT(unp); 1454 UNP_PCB_LOCK_ASSERT(unp2); 1455 1456 unp->unp_conn = NULL; 1457 switch (unp->unp_socket->so_type) { 1458 case SOCK_DGRAM: 1459 LIST_REMOVE(unp, unp_reflink); 1460 so = unp->unp_socket; 1461 SOCK_LOCK(so); 1462 so->so_state &= ~SS_ISCONNECTED; 1463 SOCK_UNLOCK(so); 1464 break; 1465 1466 case SOCK_STREAM: 1467 case SOCK_SEQPACKET: 1468 soisdisconnected(unp->unp_socket); 1469 unp2->unp_conn = NULL; 1470 soisdisconnected(unp2->unp_socket); 1471 break; 1472 } 1473} 1474 1475/* 1476 * unp_pcblist() walks the global list of struct unpcb's to generate a 1477 * pointer list, bumping the refcount on each unpcb. It then copies them out 1478 * sequentially, validating the generation number on each to see if it has 1479 * been detached. All of this is necessary because copyout() may sleep on 1480 * disk I/O. 1481 */ 1482static int 1483unp_pcblist(SYSCTL_HANDLER_ARGS) 1484{ 1485 int error, i, n; 1486 int freeunp; 1487 struct unpcb *unp, **unp_list; 1488 unp_gen_t gencnt; 1489 struct xunpgen *xug; 1490 struct unp_head *head; 1491 struct xunpcb *xu; 1492 1493 switch ((intptr_t)arg1) { 1494 case SOCK_STREAM: 1495 head = &unp_shead; 1496 break; 1497 1498 case SOCK_DGRAM: 1499 head = &unp_dhead; 1500 break; 1501 1502 case SOCK_SEQPACKET: 1503 head = &unp_sphead; 1504 break; 1505 1506 default: 1507 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1508 } 1509 1510 /* 1511 * The process of preparing the PCB list is too time-consuming and 1512 * resource-intensive to repeat twice on every request. 1513 */ 1514 if (req->oldptr == NULL) { 1515 n = unp_count; 1516 req->oldidx = 2 * (sizeof *xug) 1517 + (n + n/8) * sizeof(struct xunpcb); 1518 return (0); 1519 } 1520 1521 if (req->newptr != NULL) 1522 return (EPERM); 1523 1524 /* 1525 * OK, now we're committed to doing something. 1526 */ 1527 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1528 UNP_LIST_LOCK(); 1529 gencnt = unp_gencnt; 1530 n = unp_count; 1531 UNP_LIST_UNLOCK(); 1532 1533 xug->xug_len = sizeof *xug; 1534 xug->xug_count = n; 1535 xug->xug_gen = gencnt; 1536 xug->xug_sogen = so_gencnt; 1537 error = SYSCTL_OUT(req, xug, sizeof *xug); 1538 if (error) { 1539 free(xug, M_TEMP); 1540 return (error); 1541 } 1542 1543 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1544 1545 UNP_LIST_LOCK(); 1546 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1547 unp = LIST_NEXT(unp, unp_link)) { 1548 UNP_PCB_LOCK(unp); 1549 if (unp->unp_gencnt <= gencnt) { 1550 if (cr_cansee(req->td->td_ucred, 1551 unp->unp_socket->so_cred)) { 1552 UNP_PCB_UNLOCK(unp); 1553 continue; 1554 } 1555 unp_list[i++] = unp; 1556 unp->unp_refcount++; 1557 } 1558 UNP_PCB_UNLOCK(unp); 1559 } 1560 UNP_LIST_UNLOCK(); 1561 n = i; /* In case we lost some during malloc. */ 1562 1563 error = 0; 1564 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1565 for (i = 0; i < n; i++) { 1566 unp = unp_list[i]; 1567 UNP_PCB_LOCK(unp); 1568 unp->unp_refcount--; 1569 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1570 xu->xu_len = sizeof *xu; 1571 xu->xu_unpp = unp; 1572 /* 1573 * XXX - need more locking here to protect against 1574 * connect/disconnect races for SMP. 1575 */ 1576 if (unp->unp_addr != NULL) 1577 bcopy(unp->unp_addr, &xu->xu_addr, 1578 unp->unp_addr->sun_len); 1579 if (unp->unp_conn != NULL && 1580 unp->unp_conn->unp_addr != NULL) 1581 bcopy(unp->unp_conn->unp_addr, 1582 &xu->xu_caddr, 1583 unp->unp_conn->unp_addr->sun_len); 1584 bcopy(unp, &xu->xu_unp, sizeof *unp); 1585 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1586 UNP_PCB_UNLOCK(unp); 1587 error = SYSCTL_OUT(req, xu, sizeof *xu); 1588 } else { 1589 freeunp = (unp->unp_refcount == 0); 1590 UNP_PCB_UNLOCK(unp); 1591 if (freeunp) { 1592 UNP_PCB_LOCK_DESTROY(unp); 1593 uma_zfree(unp_zone, unp); 1594 } 1595 } 1596 } 1597 free(xu, M_TEMP); 1598 if (!error) { 1599 /* 1600 * Give the user an updated idea of our state. If the 1601 * generation differs from what we told her before, she knows 1602 * that something happened while we were processing this 1603 * request, and it might be necessary to retry. 1604 */ 1605 xug->xug_gen = unp_gencnt; 1606 xug->xug_sogen = so_gencnt; 1607 xug->xug_count = unp_count; 1608 error = SYSCTL_OUT(req, xug, sizeof *xug); 1609 } 1610 free(unp_list, M_TEMP); 1611 free(xug, M_TEMP); 1612 return (error); 1613} 1614 1615SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1616 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1617 "List of active local datagram sockets"); 1618SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1619 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1620 "List of active local stream sockets"); 1621SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1622 CTLTYPE_OPAQUE | CTLFLAG_RD, 1623 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1624 "List of active local seqpacket sockets"); 1625 1626static void 1627unp_shutdown(struct unpcb *unp) 1628{ 1629 struct unpcb *unp2; 1630 struct socket *so; 1631 1632 UNP_LINK_WLOCK_ASSERT(); 1633 UNP_PCB_LOCK_ASSERT(unp); 1634 1635 unp2 = unp->unp_conn; 1636 if ((unp->unp_socket->so_type == SOCK_STREAM || 1637 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1638 so = unp2->unp_socket; 1639 if (so != NULL) 1640 socantrcvmore(so); 1641 } 1642} 1643 1644static void 1645unp_drop(struct unpcb *unp, int errno) 1646{ 1647 struct socket *so = unp->unp_socket; 1648 struct unpcb *unp2; 1649 1650 UNP_LINK_WLOCK_ASSERT(); 1651 UNP_PCB_LOCK_ASSERT(unp); 1652 1653 so->so_error = errno; 1654 unp2 = unp->unp_conn; 1655 if (unp2 == NULL) 1656 return; 1657 UNP_PCB_LOCK(unp2); 1658 unp_disconnect(unp, unp2); 1659 UNP_PCB_UNLOCK(unp2); 1660} 1661 1662static void 1663unp_freerights(struct file **rp, int fdcount) 1664{ 1665 int i; 1666 struct file *fp; 1667 1668 for (i = 0; i < fdcount; i++) { 1669 fp = *rp; 1670 *rp++ = NULL; 1671 unp_discard(fp); 1672 } 1673} 1674 1675static int 1676unp_externalize(struct mbuf *control, struct mbuf **controlp) 1677{ 1678 struct thread *td = curthread; /* XXX */ 1679 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1680 int i; 1681 int *fdp; 1682 struct file **rp; 1683 struct file *fp; 1684 void *data; 1685 socklen_t clen = control->m_len, datalen; 1686 int error, newfds; 1687 u_int newlen; 1688 1689 UNP_LINK_UNLOCK_ASSERT(); 1690 1691 error = 0; 1692 if (controlp != NULL) /* controlp == NULL => free control messages */ 1693 *controlp = NULL; 1694 while (cm != NULL) { 1695 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1696 error = EINVAL; 1697 break; 1698 } 1699 data = CMSG_DATA(cm); 1700 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1701 if (cm->cmsg_level == SOL_SOCKET 1702 && cm->cmsg_type == SCM_RIGHTS) { 1703 newfds = datalen / sizeof(struct file *); 1704 rp = data; 1705 1706 /* If we're not outputting the descriptors free them. */ 1707 if (error || controlp == NULL) { 1708 unp_freerights(rp, newfds); 1709 goto next; 1710 } 1711 FILEDESC_XLOCK(td->td_proc->p_fd); 1712 /* 1713 * Now change each pointer to an fd in the global 1714 * table to an integer that is the index to the local 1715 * fd table entry that we set up to point to the 1716 * global one we are transferring. 1717 */ 1718 newlen = newfds * sizeof(int); 1719 *controlp = sbcreatecontrol(NULL, newlen, 1720 SCM_RIGHTS, SOL_SOCKET); 1721 if (*controlp == NULL) { 1722 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1723 error = E2BIG; 1724 unp_freerights(rp, newfds); 1725 goto next; 1726 } 1727 1728 fdp = (int *) 1729 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1730 if (fdallocn(td, 0, fdp, newfds) != 0) { 1731 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1732 error = EMSGSIZE; 1733 unp_freerights(rp, newfds); 1734 m_freem(*controlp); 1735 *controlp = NULL; 1736 goto next; 1737 } 1738 for (i = 0; i < newfds; i++) { 1739 fp = *rp++; 1740 td->td_proc->p_fd->fd_ofiles[fdp[i]] = fp; 1741 unp_externalize_fp(fp); 1742 } 1743 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1744 } else { 1745 /* We can just copy anything else across. */ 1746 if (error || controlp == NULL) 1747 goto next; 1748 *controlp = sbcreatecontrol(NULL, datalen, 1749 cm->cmsg_type, cm->cmsg_level); 1750 if (*controlp == NULL) { 1751 error = ENOBUFS; 1752 goto next; 1753 } 1754 bcopy(data, 1755 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1756 datalen); 1757 } 1758 controlp = &(*controlp)->m_next; 1759 1760next: 1761 if (CMSG_SPACE(datalen) < clen) { 1762 clen -= CMSG_SPACE(datalen); 1763 cm = (struct cmsghdr *) 1764 ((caddr_t)cm + CMSG_SPACE(datalen)); 1765 } else { 1766 clen = 0; 1767 cm = NULL; 1768 } 1769 } 1770 1771 m_freem(control); 1772 return (error); 1773} 1774 1775static void 1776unp_zone_change(void *tag) 1777{ 1778 1779 uma_zone_set_max(unp_zone, maxsockets); 1780} 1781 1782static void 1783unp_init(void) 1784{ 1785 1786#ifdef VIMAGE 1787 if (!IS_DEFAULT_VNET(curvnet)) 1788 return; 1789#endif 1790 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1791 NULL, NULL, UMA_ALIGN_PTR, 0); 1792 if (unp_zone == NULL) 1793 panic("unp_init"); 1794 uma_zone_set_max(unp_zone, maxsockets); 1795 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1796 NULL, EVENTHANDLER_PRI_ANY); 1797 LIST_INIT(&unp_dhead); 1798 LIST_INIT(&unp_shead); 1799 LIST_INIT(&unp_sphead); 1800 SLIST_INIT(&unp_defers); 1801 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 1802 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1803 UNP_LINK_LOCK_INIT(); 1804 UNP_LIST_LOCK_INIT(); 1805 UNP_DEFERRED_LOCK_INIT(); 1806} 1807 1808static int 1809unp_internalize(struct mbuf **controlp, struct thread *td) 1810{ 1811 struct mbuf *control = *controlp; 1812 struct proc *p = td->td_proc; 1813 struct filedesc *fdescp = p->p_fd; 1814 struct bintime *bt; 1815 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1816 struct cmsgcred *cmcred; 1817 struct file **rp; 1818 struct file *fp; 1819 struct timeval *tv; 1820 int i, fd, *fdp; 1821 void *data; 1822 socklen_t clen = control->m_len, datalen; 1823 int error, oldfds; 1824 u_int newlen; 1825 1826 UNP_LINK_UNLOCK_ASSERT(); 1827 1828 error = 0; 1829 *controlp = NULL; 1830 while (cm != NULL) { 1831 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1832 || cm->cmsg_len > clen) { 1833 error = EINVAL; 1834 goto out; 1835 } 1836 data = CMSG_DATA(cm); 1837 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1838 1839 switch (cm->cmsg_type) { 1840 /* 1841 * Fill in credential information. 1842 */ 1843 case SCM_CREDS: 1844 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1845 SCM_CREDS, SOL_SOCKET); 1846 if (*controlp == NULL) { 1847 error = ENOBUFS; 1848 goto out; 1849 } 1850 cmcred = (struct cmsgcred *) 1851 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1852 cmcred->cmcred_pid = p->p_pid; 1853 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1854 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1855 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1856 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1857 CMGROUP_MAX); 1858 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1859 cmcred->cmcred_groups[i] = 1860 td->td_ucred->cr_groups[i]; 1861 break; 1862 1863 case SCM_RIGHTS: 1864 oldfds = datalen / sizeof (int); 1865 /* 1866 * Check that all the FDs passed in refer to legal 1867 * files. If not, reject the entire operation. 1868 */ 1869 fdp = data; 1870 FILEDESC_SLOCK(fdescp); 1871 for (i = 0; i < oldfds; i++) { 1872 fd = *fdp++; 1873 if ((unsigned)fd >= fdescp->fd_nfiles || 1874 fdescp->fd_ofiles[fd] == NULL) { 1875 FILEDESC_SUNLOCK(fdescp); 1876 error = EBADF; 1877 goto out; 1878 } 1879 fp = fdescp->fd_ofiles[fd]; 1880 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1881 FILEDESC_SUNLOCK(fdescp); 1882 error = EOPNOTSUPP; 1883 goto out; 1884 } 1885 1886 } 1887 1888 /* 1889 * Now replace the integer FDs with pointers to the 1890 * associated global file table entry.. 1891 */ 1892 newlen = oldfds * sizeof(struct file *); 1893 *controlp = sbcreatecontrol(NULL, newlen, 1894 SCM_RIGHTS, SOL_SOCKET); 1895 if (*controlp == NULL) { 1896 FILEDESC_SUNLOCK(fdescp); 1897 error = E2BIG; 1898 goto out; 1899 } 1900 fdp = data; 1901 rp = (struct file **) 1902 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1903 for (i = 0; i < oldfds; i++) { 1904 fp = fdescp->fd_ofiles[*fdp++]; 1905 *rp++ = fp; 1906 unp_internalize_fp(fp); 1907 } 1908 FILEDESC_SUNLOCK(fdescp); 1909 break; 1910 1911 case SCM_TIMESTAMP: 1912 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1913 SCM_TIMESTAMP, SOL_SOCKET); 1914 if (*controlp == NULL) { 1915 error = ENOBUFS; 1916 goto out; 1917 } 1918 tv = (struct timeval *) 1919 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1920 microtime(tv); 1921 break; 1922 1923 case SCM_BINTIME: 1924 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 1925 SCM_BINTIME, SOL_SOCKET); 1926 if (*controlp == NULL) { 1927 error = ENOBUFS; 1928 goto out; 1929 } 1930 bt = (struct bintime *) 1931 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1932 bintime(bt); 1933 break; 1934 1935 default: 1936 error = EINVAL; 1937 goto out; 1938 } 1939 1940 controlp = &(*controlp)->m_next; 1941 if (CMSG_SPACE(datalen) < clen) { 1942 clen -= CMSG_SPACE(datalen); 1943 cm = (struct cmsghdr *) 1944 ((caddr_t)cm + CMSG_SPACE(datalen)); 1945 } else { 1946 clen = 0; 1947 cm = NULL; 1948 } 1949 } 1950 1951out: 1952 m_freem(control); 1953 return (error); 1954} 1955 1956static struct mbuf * 1957unp_addsockcred(struct thread *td, struct mbuf *control) 1958{ 1959 struct mbuf *m, *n, *n_prev; 1960 struct sockcred *sc; 1961 const struct cmsghdr *cm; 1962 int ngroups; 1963 int i; 1964 1965 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1966 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1967 if (m == NULL) 1968 return (control); 1969 1970 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1971 sc->sc_uid = td->td_ucred->cr_ruid; 1972 sc->sc_euid = td->td_ucred->cr_uid; 1973 sc->sc_gid = td->td_ucred->cr_rgid; 1974 sc->sc_egid = td->td_ucred->cr_gid; 1975 sc->sc_ngroups = ngroups; 1976 for (i = 0; i < sc->sc_ngroups; i++) 1977 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1978 1979 /* 1980 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1981 * created SCM_CREDS control message (struct sockcred) has another 1982 * format. 1983 */ 1984 if (control != NULL) 1985 for (n = control, n_prev = NULL; n != NULL;) { 1986 cm = mtod(n, struct cmsghdr *); 1987 if (cm->cmsg_level == SOL_SOCKET && 1988 cm->cmsg_type == SCM_CREDS) { 1989 if (n_prev == NULL) 1990 control = n->m_next; 1991 else 1992 n_prev->m_next = n->m_next; 1993 n = m_free(n); 1994 } else { 1995 n_prev = n; 1996 n = n->m_next; 1997 } 1998 } 1999 2000 /* Prepend it to the head. */ 2001 m->m_next = control; 2002 return (m); 2003} 2004 2005static struct unpcb * 2006fptounp(struct file *fp) 2007{ 2008 struct socket *so; 2009 2010 if (fp->f_type != DTYPE_SOCKET) 2011 return (NULL); 2012 if ((so = fp->f_data) == NULL) 2013 return (NULL); 2014 if (so->so_proto->pr_domain != &localdomain) 2015 return (NULL); 2016 return sotounpcb(so); 2017} 2018 2019static void 2020unp_discard(struct file *fp) 2021{ 2022 struct unp_defer *dr; 2023 2024 if (unp_externalize_fp(fp)) { 2025 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2026 dr->ud_fp = fp; 2027 UNP_DEFERRED_LOCK(); 2028 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2029 UNP_DEFERRED_UNLOCK(); 2030 atomic_add_int(&unp_defers_count, 1); 2031 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2032 } else 2033 (void) closef(fp, (struct thread *)NULL); 2034} 2035 2036static void 2037unp_process_defers(void *arg __unused, int pending) 2038{ 2039 struct unp_defer *dr; 2040 SLIST_HEAD(, unp_defer) drl; 2041 int count; 2042 2043 SLIST_INIT(&drl); 2044 for (;;) { 2045 UNP_DEFERRED_LOCK(); 2046 if (SLIST_FIRST(&unp_defers) == NULL) { 2047 UNP_DEFERRED_UNLOCK(); 2048 break; 2049 } 2050 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2051 UNP_DEFERRED_UNLOCK(); 2052 count = 0; 2053 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2054 SLIST_REMOVE_HEAD(&drl, ud_link); 2055 closef(dr->ud_fp, NULL); 2056 free(dr, M_TEMP); 2057 count++; 2058 } 2059 atomic_add_int(&unp_defers_count, -count); 2060 } 2061} 2062 2063static void 2064unp_internalize_fp(struct file *fp) 2065{ 2066 struct unpcb *unp; 2067 2068 UNP_LINK_WLOCK(); 2069 if ((unp = fptounp(fp)) != NULL) { 2070 unp->unp_file = fp; 2071 unp->unp_msgcount++; 2072 } 2073 fhold(fp); 2074 unp_rights++; 2075 UNP_LINK_WUNLOCK(); 2076} 2077 2078static int 2079unp_externalize_fp(struct file *fp) 2080{ 2081 struct unpcb *unp; 2082 int ret; 2083 2084 UNP_LINK_WLOCK(); 2085 if ((unp = fptounp(fp)) != NULL) { 2086 unp->unp_msgcount--; 2087 ret = 1; 2088 } else 2089 ret = 0; 2090 unp_rights--; 2091 UNP_LINK_WUNLOCK(); 2092 return (ret); 2093} 2094 2095/* 2096 * unp_defer indicates whether additional work has been defered for a future 2097 * pass through unp_gc(). It is thread local and does not require explicit 2098 * synchronization. 2099 */ 2100static int unp_marked; 2101static int unp_unreachable; 2102 2103static void 2104unp_accessable(struct file *fp) 2105{ 2106 struct unpcb *unp; 2107 2108 if ((unp = fptounp(fp)) == NULL) 2109 return; 2110 if (unp->unp_gcflag & UNPGC_REF) 2111 return; 2112 unp->unp_gcflag &= ~UNPGC_DEAD; 2113 unp->unp_gcflag |= UNPGC_REF; 2114 unp_marked++; 2115} 2116 2117static void 2118unp_gc_process(struct unpcb *unp) 2119{ 2120 struct socket *soa; 2121 struct socket *so; 2122 struct file *fp; 2123 2124 /* Already processed. */ 2125 if (unp->unp_gcflag & UNPGC_SCANNED) 2126 return; 2127 fp = unp->unp_file; 2128 2129 /* 2130 * Check for a socket potentially in a cycle. It must be in a 2131 * queue as indicated by msgcount, and this must equal the file 2132 * reference count. Note that when msgcount is 0 the file is NULL. 2133 */ 2134 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2135 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2136 unp->unp_gcflag |= UNPGC_DEAD; 2137 unp_unreachable++; 2138 return; 2139 } 2140 2141 /* 2142 * Mark all sockets we reference with RIGHTS. 2143 */ 2144 so = unp->unp_socket; 2145 SOCKBUF_LOCK(&so->so_rcv); 2146 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2147 SOCKBUF_UNLOCK(&so->so_rcv); 2148 2149 /* 2150 * Mark all sockets in our accept queue. 2151 */ 2152 ACCEPT_LOCK(); 2153 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2154 SOCKBUF_LOCK(&soa->so_rcv); 2155 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2156 SOCKBUF_UNLOCK(&soa->so_rcv); 2157 } 2158 ACCEPT_UNLOCK(); 2159 unp->unp_gcflag |= UNPGC_SCANNED; 2160} 2161 2162static int unp_recycled; 2163SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2164 "Number of unreachable sockets claimed by the garbage collector."); 2165 2166static int unp_taskcount; 2167SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2168 "Number of times the garbage collector has run."); 2169 2170static void 2171unp_gc(__unused void *arg, int pending) 2172{ 2173 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2174 NULL }; 2175 struct unp_head **head; 2176 struct file *f, **unref; 2177 struct unpcb *unp; 2178 int i, total; 2179 2180 unp_taskcount++; 2181 UNP_LIST_LOCK(); 2182 /* 2183 * First clear all gc flags from previous runs. 2184 */ 2185 for (head = heads; *head != NULL; head++) 2186 LIST_FOREACH(unp, *head, unp_link) 2187 unp->unp_gcflag = 0; 2188 2189 /* 2190 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2191 * is reachable all of the sockets it references are reachable. 2192 * Stop the scan once we do a complete loop without discovering 2193 * a new reachable socket. 2194 */ 2195 do { 2196 unp_unreachable = 0; 2197 unp_marked = 0; 2198 for (head = heads; *head != NULL; head++) 2199 LIST_FOREACH(unp, *head, unp_link) 2200 unp_gc_process(unp); 2201 } while (unp_marked); 2202 UNP_LIST_UNLOCK(); 2203 if (unp_unreachable == 0) 2204 return; 2205 2206 /* 2207 * Allocate space for a local list of dead unpcbs. 2208 */ 2209 unref = malloc(unp_unreachable * sizeof(struct file *), 2210 M_TEMP, M_WAITOK); 2211 2212 /* 2213 * Iterate looking for sockets which have been specifically marked 2214 * as as unreachable and store them locally. 2215 */ 2216 UNP_LINK_RLOCK(); 2217 UNP_LIST_LOCK(); 2218 for (total = 0, head = heads; *head != NULL; head++) 2219 LIST_FOREACH(unp, *head, unp_link) 2220 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2221 f = unp->unp_file; 2222 if (unp->unp_msgcount == 0 || f == NULL || 2223 f->f_count != unp->unp_msgcount) 2224 continue; 2225 unref[total++] = f; 2226 fhold(f); 2227 KASSERT(total <= unp_unreachable, 2228 ("unp_gc: incorrect unreachable count.")); 2229 } 2230 UNP_LIST_UNLOCK(); 2231 UNP_LINK_RUNLOCK(); 2232 2233 /* 2234 * Now flush all sockets, free'ing rights. This will free the 2235 * struct files associated with these sockets but leave each socket 2236 * with one remaining ref. 2237 */ 2238 for (i = 0; i < total; i++) { 2239 struct socket *so; 2240 2241 so = unref[i]->f_data; 2242 CURVNET_SET(so->so_vnet); 2243 sorflush(so); 2244 CURVNET_RESTORE(); 2245 } 2246 2247 /* 2248 * And finally release the sockets so they can be reclaimed. 2249 */ 2250 for (i = 0; i < total; i++) 2251 fdrop(unref[i], NULL); 2252 unp_recycled += total; 2253 free(unref, M_TEMP); 2254} 2255 2256static void 2257unp_dispose(struct mbuf *m) 2258{ 2259 2260 if (m) 2261 unp_scan(m, unp_discard); 2262} 2263 2264static void 2265unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2266{ 2267 struct mbuf *m; 2268 struct file **rp; 2269 struct cmsghdr *cm; 2270 void *data; 2271 int i; 2272 socklen_t clen, datalen; 2273 int qfds; 2274 2275 while (m0 != NULL) { 2276 for (m = m0; m; m = m->m_next) { 2277 if (m->m_type != MT_CONTROL) 2278 continue; 2279 2280 cm = mtod(m, struct cmsghdr *); 2281 clen = m->m_len; 2282 2283 while (cm != NULL) { 2284 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2285 break; 2286 2287 data = CMSG_DATA(cm); 2288 datalen = (caddr_t)cm + cm->cmsg_len 2289 - (caddr_t)data; 2290 2291 if (cm->cmsg_level == SOL_SOCKET && 2292 cm->cmsg_type == SCM_RIGHTS) { 2293 qfds = datalen / sizeof (struct file *); 2294 rp = data; 2295 for (i = 0; i < qfds; i++) 2296 (*op)(*rp++); 2297 } 2298 2299 if (CMSG_SPACE(datalen) < clen) { 2300 clen -= CMSG_SPACE(datalen); 2301 cm = (struct cmsghdr *) 2302 ((caddr_t)cm + CMSG_SPACE(datalen)); 2303 } else { 2304 clen = 0; 2305 cm = NULL; 2306 } 2307 } 2308 } 2309 m0 = m0->m_act; 2310 } 2311} 2312 2313/* 2314 * A helper function called by VFS before socket-type vnode reclamation. 2315 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2316 * use count. 2317 */ 2318void 2319vfs_unp_reclaim(struct vnode *vp) 2320{ 2321 struct socket *so; 2322 struct unpcb *unp; 2323 int active; 2324 2325 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2326 KASSERT(vp->v_type == VSOCK, 2327 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2328 2329 active = 0; 2330 UNP_LINK_WLOCK(); 2331 VOP_UNP_CONNECT(vp, &so); 2332 if (so == NULL) 2333 goto done; 2334 unp = sotounpcb(so); 2335 if (unp == NULL) 2336 goto done; 2337 UNP_PCB_LOCK(unp); 2338 if (unp->unp_vnode == vp) { 2339 VOP_UNP_DETACH(vp); 2340 unp->unp_vnode = NULL; 2341 active = 1; 2342 } 2343 UNP_PCB_UNLOCK(unp); 2344done: 2345 UNP_LINK_WUNLOCK(); 2346 if (active) 2347 vunref(vp); 2348} 2349 2350#ifdef DDB 2351static void 2352db_print_indent(int indent) 2353{ 2354 int i; 2355 2356 for (i = 0; i < indent; i++) 2357 db_printf(" "); 2358} 2359 2360static void 2361db_print_unpflags(int unp_flags) 2362{ 2363 int comma; 2364 2365 comma = 0; 2366 if (unp_flags & UNP_HAVEPC) { 2367 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2368 comma = 1; 2369 } 2370 if (unp_flags & UNP_HAVEPCCACHED) { 2371 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2372 comma = 1; 2373 } 2374 if (unp_flags & UNP_WANTCRED) { 2375 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2376 comma = 1; 2377 } 2378 if (unp_flags & UNP_CONNWAIT) { 2379 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2380 comma = 1; 2381 } 2382 if (unp_flags & UNP_CONNECTING) { 2383 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2384 comma = 1; 2385 } 2386 if (unp_flags & UNP_BINDING) { 2387 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2388 comma = 1; 2389 } 2390} 2391 2392static void 2393db_print_xucred(int indent, struct xucred *xu) 2394{ 2395 int comma, i; 2396 2397 db_print_indent(indent); 2398 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2399 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2400 db_print_indent(indent); 2401 db_printf("cr_groups: "); 2402 comma = 0; 2403 for (i = 0; i < xu->cr_ngroups; i++) { 2404 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2405 comma = 1; 2406 } 2407 db_printf("\n"); 2408} 2409 2410static void 2411db_print_unprefs(int indent, struct unp_head *uh) 2412{ 2413 struct unpcb *unp; 2414 int counter; 2415 2416 counter = 0; 2417 LIST_FOREACH(unp, uh, unp_reflink) { 2418 if (counter % 4 == 0) 2419 db_print_indent(indent); 2420 db_printf("%p ", unp); 2421 if (counter % 4 == 3) 2422 db_printf("\n"); 2423 counter++; 2424 } 2425 if (counter != 0 && counter % 4 != 0) 2426 db_printf("\n"); 2427} 2428 2429DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2430{ 2431 struct unpcb *unp; 2432 2433 if (!have_addr) { 2434 db_printf("usage: show unpcb <addr>\n"); 2435 return; 2436 } 2437 unp = (struct unpcb *)addr; 2438 2439 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2440 unp->unp_vnode); 2441 2442 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2443 unp->unp_conn); 2444 2445 db_printf("unp_refs:\n"); 2446 db_print_unprefs(2, &unp->unp_refs); 2447 2448 /* XXXRW: Would be nice to print the full address, if any. */ 2449 db_printf("unp_addr: %p\n", unp->unp_addr); 2450 2451 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2452 unp->unp_cc, unp->unp_mbcnt, 2453 (unsigned long long)unp->unp_gencnt); 2454 2455 db_printf("unp_flags: %x (", unp->unp_flags); 2456 db_print_unpflags(unp->unp_flags); 2457 db_printf(")\n"); 2458 2459 db_printf("unp_peercred:\n"); 2460 db_print_xucred(2, &unp->unp_peercred); 2461 2462 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2463} 2464#endif 2465